Related Experiment Videos
Predicting drug approvals: The Novartis data science and artificial intelligence challenge
Kien Wei Siah1,2, Nicholas W Kelley3, Steffen Ballerstedt3
1Laboratory for Financial Engineering, Sloan School of Management, Massachusetts Institute of Technology, Cambridge, MA 02142, USA.
Patterns (New York, N.Y.)
|August 25, 2021
Summary
Novartis
Area of Science:
- Drug development
- Machine learning
- Artificial intelligence
Background:
- Academic research from MIT utilized Informa data to predict drug development outcomes.
- A collaboration between academia and industry was initiated to enhance predictive modeling.
- Drug development success prediction remains a significant challenge in pharmaceutical research.
Purpose of the Study:
- To develop advanced machine learning models for predicting drug development success.
- To leverage cross-functional expertise within Novartis to improve predictive accuracy.
- To identify novel features and data driving drug development outcomes.
Main Methods:
- An in-house data science and artificial intelligence challenge was conducted.
- Over 50 cross-functional teams from 25 global Novartis offices participated.
- State-of-the-art machine learning algorithms and new data features were employed.
Main Results:
- Two winning models significantly outperformed the baseline MIT model (AUC 0.88 and 0.84 vs. 0.78).
- Predictive models incorporated newly identified features and data sources.
- Validated previously identified variables associated with drug approval.
Conclusions:
- The collaboration successfully enhanced machine learning models for predicting drug development outcomes.
- The challenge provided new insights into factors influencing drug development success and failure.
- Industry-academia partnerships can drive innovation in pharmaceutical AI.